This directory contains code for estimating the DeepIV model

1) The infos_feature_new.csv contains the data used for estimating the DeepIV model in the following form:

Hashtag_name video_id impr hashtag_age if_trending ranking if_spon video_age if_verified num_hashtag num_trending if_fyp following num_video follower second_nn_avg_heart topics entertaining informational socializing shares likes comments feature_256

2) The find_iv.py contains the code that identifies the second nearest neighbor of the focal video and saves its creator's average historical hearts as the IV 

3) The deepiv.py contains the code that calculates the average treatment effect of followers, the heterogeneous effects by engagement tactics, topics, and sponsored campaigns, and a profit maximization case study for Walmart. Overall, it should take around 6-7 weeks to finish in a gpu-enabled high performance computing environment. 